基于卡尔曼预测的视频目标实时跟踪

在视频目标跟踪中,搜索区间大小直接影响着跟踪的速度和效率.现采用图象坐标系下的卡尔曼滤波预测来指导跟踪,设计了卡尔曼滤波器,对被跟踪目标的运动参数(位置,速度,加速度)进行滤波预测.缩小了搜索区间(减至为匹配模型的大小),提高了系统的实时性和跟踪精度.实验结果表明:在对大运动和大运动大机动两种视频目标的跟踪中,耗时仅为全局搜索的1/4,精度可达3像素以内.

A Real-Time Tracking for Video Target Based on Kalman Prediction

For video target tracking,searching area is focused with great affect on tracking at high speed and efficienctly.In this paper,Kalman filter prediction is introduced to guide target tracking in video under imagine reference frame.And a Kalman filter is designed to flit and predict motion parameter (position,velocity,acceleration) of tracked target.By this method,searching area is reduced(reduced to the area of a matched model),response time and accuracy of tracking are improved.The results show that tracking accuracy is within 3 pixels and time consuming is only 1/4 for whole area searching in the fast maneuvering target tracking.